FairSwiRL: fair semi-supervised classification with representation learning

Abstract Semi-supervised learning has shown its potential in many real-world applications where only few labeled examples are available. However, when some fairness constraints need to be satisfied, semi-supervised classification models often struggle as they required cope with the lack of sufficient information for predicting target variable while forgetting relationships any sensitive and potentially discriminatory attribute. To address this issue, we propose a fair representation architecture that leads accurate results even very challenging scenarios (but biased) instances. We show experimentally our model can easily adopted general settings, learned representations may employed train supervised classifier. Moreover, applied several synthetic datasets, method is competitive state-of-the-art approaches.